Measured, not asserted

Data quality

Every number below is read from a committed engine artifact — this page cannot say anything the runs didn't measure. Each stat links its reproducible evidence; each figure carries the date it was measured. The standing promises behind these guards live in DATA_SLA.md.

North star — full-surface audit ok-rate

Hundreds of cold, natural probes across every retrieval surface, graded against ground truth. The one number the whole engine system optimizes.

latest (2026-07-11)
82%
163/198 probes ok · target ≥85%

SCF funding cross-check

No project overstates or understates SCF membership vs the fund's own directory.

as of 2026-07-11
0 / 0
overstated / understated memberships
  • 317 matched records checked against 547 SCF projects
  • 271 per-round claims verified
  • 13 round-level overclaims found by this run (queued as its fix wave)

Contract honesty probe

Documented params do something, undocumented values are rejected — the contract a stranger hits behaves as written.

as of 2026-07-11
229
params + fields probed across 24 operations
  • 1 silent param(s), 4 invalid-accepted param(s) at baseline (spec 1.7.16)
  • Re-run on every deploy (post-deploy-eval workflow)

Code-depth calibration

Repo depth grades agree with independent code analysis where both exist.

as of 2026-07-10
100%
agreement on 3 co-graded repos (83 sampled)
  • 0 disagreements
  • 80 sampled repos had no independent index to compare (small-n baseline — grows as coverage does)

Consumer interlock (Raven)

The #1 consumer's discovery index tracks our contract — checked from OUR side too, with grace for their re-baseline cadence.

as of 2026-07-21
20 / 22
operations discoverable in the consumer catalog
  • 2 op(s) lagging within the 10-day re-baseline grace window (expected)
  • 0 op(s) missing beyond grace
  • contract 1.8.14 at measurement

Recall floors (Engine A)

Generated known-item probes per bucket stay above their red-line floors — recall can't silently erode.

as of latest weekly run
7/8
buckets at/above floor · 2,135 probes
  • R-SYM at 64.3% vs floor 75% — open red

Real-demand OK-rate (Engine D)

The queries real consumers actually sent are replayed live — a miss on real demand outranks any synthetic finding.

as of latest weekly run
88%
top 250 of 4,084 distinct real queries
  • 5,313 real-consumer calls in the 14-day window
  • 30 queries missing today — the standing fix queue

Golden retrieval eval

Known-true questions (answer key derived from the canonical directory) keep passing after every ship.

as of latest weekly run
51/51
golden questions passing (scored)
  • full pass
  • 2 N/A by design (live-source questions the static corpus doesn't answer)
  • re-run on every production deploy + weekly

Corpus hygiene (S5–S8)

The research corpus stays clean — junk URLs, broken titles, staleness and mirror drift are swept weekly.

as of latest weekly run
2/5
sweeps clean · 9,808 chunks / 1,844 docs
  • s5 junk urls: 1 flagged — queued
  • s6 bad titles: 69 flagged — queued
  • s8 mirrors: 127 flagged — queued

Consumer path (through Raven)

The canonical questions answer correctly through the REAL Raven gateway — routing, our op, the response envelope and coaching — not just our direct API.

as of 2026-07-23
44/44
golden Qs answered through the live gateway · 100%
  • every gradeable golden question answers correctly through Raven (44 checked)

Improvement ledger

Every quality detector's findings land in one tracked backlog by surface. A backlog is fine; a HIGH-severity finding neglected past 30 days is the failure — that's this row's red line.

as of 2026-08-17
307 open
18 high · 0 in a wave · 24% closed
  • open by surface: retrieval 246 · directory 23 · contract 14 · consumer 14 · corpus 6 · code 4 · scf 0
  • 404 tracked · 0 in-wave · 7 verified · 90 auto-cleared (detector stopped flagging)
  • no high-severity finding neglected past 30 days
  • ⚠ raven-routing (25d, 14 open), raven-loop (25d, 1 open) — evidence past the 10d window; 15 open finding(s) unconfirmed

How this page works: engine runs commit their raw JSON evidence to improvements/ and this page statically renders those files — a number here can only change when a new dated artifact lands. The consumer-side interlock conventions (spec-as-discovery-index, version handshake, cadence contract) are specified in docs/interlock-spec.md.